Using Model Atomic Spin Systems to Learn About in materia Computing

The quest to implement machine learning algorithms in hardware has focused on combining various materials, each mimicking a computational primitive, to create device functionality. These endeavors have led to the beautiful development of dedicated hardware that, working in combination with software, can perform pattern recognition tasks. Ultimately, these piecewise approaches limit functionality and efficiency, while complicating scaling and on-chip learning, necessitating new approaches linking physical phenomena to machine learning models. Likewise, this raises the question if there are new machine learning algorithms to be discovered, utilizing the particular properties of quantum properties of matter where there are no obvious links to established models. Here, I will discuss the first steps toward a new paradigm in computing, routed in fundamentals studies based on the idea of letting the physics do the work. I will introduce the concept of an atomic orbital memory and how coupling leads to tunable multi-well landscapes. I will discuss how the ensuing stochastic dynamics mimics the perennial model in machine learning, the Boltzmann machine. In this discussion, I will review the emergence of multiple and separable time scales, an adaption of long-term potentiation in biological matter, which serves the basis for self-adaption. I will also discuss more recent develops moving toward local gate control of stochastic behavior, AC response, and finally new types of orbital memory. I will conclude with an outlook on concepts that go beyond the current neuromorphic paradigm, combining concepts related to quantum coherent and quantum technologies.